Economic Dispatch Solution Using A Genetic Algorithm Based on Arithmetic Crossover

T. Yalcinoz, Member, IEEE, H. Altun, Member, IEEE and M. Uzam, Member, IEEE
have presented a genetic algorithm for solving economic dispatch problem. The proposed method can take into account network losses, ramp rate limits and valve point zone. A fuzzy logic controlled genetic algorithm has been applied to environmental/economic dispatch by Song et. al. [2]. Song and Chou [3] have proposed a hybrid genetic algorithm that is combination strategy involving local search algorithms and genetic algorithm. The validity of a fuzzy logic controlled genetic algorithm [2] and a hybrid genetic algorithm [3] is illustrated on the economic dispatch problem with a six unit system. In the above papers, the binary representation was applied to economic dispatch problems. In this paper, elitism, arithmetic crossover which defines a linear combination of two chromosome [11] and mutation are used in the genetic algorithm to generate successive sets of possible operating policies. II. ECONOMIC DISPATCH PROBLEM The economic dispatch problem is to simultaneously minimize the overall cost rate and meet the load demand of a power system. The power system model consists of n generating units already connected to the system. The economic dispatch is to determine the optimal share of load demand for each unit in the range of 3 to 5 minutes [12]. The economic dispatch problem can be expressed as

AbstractIn this paper, a new genetic approach based on arithmetic crossover for solving the economic dispatch problem is proposed. Elitism, arithmetic crossover and mutation are used in the genetic algorithm to generate successive sets of possible operating policies. The proposed technique improves the quality of the solution. The new genetic approach is compared with an improved Hopfield NN approach (IHN) [1], a fuzzy logic controlled genetic algorithm (FLCGA) [2], an advance engineered-conditioning genetic approach (AECGA) [3] and an advance Hopfield NN approach (AHNN) [4]. Index Terms--Economic dispatch, genetic algorithm, arithmetic crossover.

I. INTRODUCTION

ITH the development of modern power systems, economic dispatch (ED) problem has received an increasing attention. The economic dispatch aims to minimize the fuel cost while provides consumers with adequate and secure electricity. The literature of the ED problem and its solution methods are surveyed in [5] and [6]. However, it is realized that conventional techniques become very complicated when dealing with increasingly complex dispatch problems, and are further limited by their lack of robustness and efficiency in a number of practical applications. Thus developing a reliable, fast and efficient algorithm is still an active area in power systems. Recently, a global optimization technique known as genetic algorithm (GA) which is a kind of the probabilistic heuristic algorithm has been studied to solve the power optimization problems. The GA may find the several sub-optimum solutions within a realistic computation time. Even if there is no guaranty that the GA may find the globally optimal solutions in a finite time. Genetic algorithm is an attractive tool for economic dispatch problems. Sheble, et. al. [7,8] presented papers where GA solves the economic dispatch. In the papers, an example with three units was explored. Bakirtzis et al. [9] have proposed a simple genetic algorithm solution to the economic dispatch problem. The operation cost obtained from GA was 0.5 % higher than the optimum operation cost. Chang and Chen [10]
This work is supported by the Research Fund of University of Nigde under the project number of FBE 2001/04. T. Yalcinoz, H. Altun and M. Uzam are adjuncts to Department of Electrical and Electronic Engineering, Nigde University, Nigde 51100, Turkey. (e-mail: tyalcinoz@ieee.org).

Min Fi ( Pi )
Pi i =1

(1)

Fi ( Pi ) =(ai + bi Pi + ci Pi 2 )
where ai, bi and ci are the cost coefficients of the i-th generator and n is the number of generators committed to the operating system. Pi is the power output of the i-th generator. The economic dispatch problem subjects to the following constraints:

Pmin,i Pi Pmax,i

for i = 1, . . . , n

(2) (3) (4)

P D L = 0
i i =1

where

L = Bi Pi 2
i =1

where D is the load demand and L represents the transmission losses. B represents coefficients of transmission losses. Pmin,i and Pmax,i are minimum and maximum generation output of the i-th generator. III. GENETIC ALGORITHM Genetic algorithms are search algorithms based on the process of biological evolution. In genetic algorithms, the mechanics of natural selection and genetics are emulated artificially. The search for a global optimum to an optimization problem is conducted by moving from an old population of individuals to a new population using genetics-like operators. Each individual represents a candidate to the optimization solution. An individual is modeled as a fixed length string of symbols, usually taken from the binary alphabet. An evaluation function, called fitness function, assigns a fitness value to each individual within the population. This fitness value is measure for the quality of an individual. The basic optimization procedure involves nothing more than processing highly fit individuals in order to produce better individuals as the search progresses. A typical genetic algorithm cycle involves four major processes of fitness evaluation, selection, recombination and creation of a new population. Although the binary representation is usually applied to power optimization problems, in this paper, we use the real valued representation scheme for solution. The use of real valued representation in the GA is claimed by Wright [13] to offer a number of advantages in numerical function optimization over binary encoding. Efficiency of the GA is increased as there is no need to convert chromosomes to the binary type; less memory is required as efficient floating-point internal computer representations can be used directly; there is no loss in precision by discretisation to binary or other values; and there is greater freedom to use different genetic operators [11]. For the real valued representation, the k-th chromosome Ck can be defined as follows: Ck = [ Pk1, Pk2, . . . , Pkn] k = 1, 2, . . . , popsize (5)

C igen +1 = a C igen + (1 a ) C jgen

+1 C gen = (1 a ) C igen + a C jgen j

(6)

where a is a random number in range of [0, 1]. The mutation operator is used to inject new genetic material into the population and it is applied to each new structure individually. A given mutation involves randomly altering each gene with a small probability. We generate a random real value which makes a random change in the m-th element selected randomly of the chromosome. The objective function is used to provide a measure of how individuals have performed in the problem domain. In the case of a minimization problem, the most fit individuals will have the lowest value of the associated objective function. The fitness function is normally used to transform the objective function value into a measure of relative fitness. The fitness function is defined as

Fit ( x ) = g (f ( x ))

(7)

where f(x) is the objective function, g transforms the value of the objective function to non-negative number. An elitist which GA search is used guarantees that the best solution so far obtained in the search is retained and used in the following generation, and thereby ensuring no good solution already found can be lost in the search process. IV. ECONOMIC DISPATCH USING GENETIC ALGORITHM The success of the genetic algorithm strongly depends on the problem mapping which involves the transformation of the problem solution to a chromosome representation and the design of the fitness function as assess the quality of a solution. Each chromosome within the population represents a candidate solution. A chromosome must represent a generation scheduling in order to solve the economic dispatch problem by using a genetic algorithm approach. In the economic dispatch problem, the unit power output is used as the main decision variable, and each units loading range is represented by a real number. The representation takes care of the unit minimum and maximum loading limits since the real representation is made to cover only the values between the limits. The main objective of the economic dispatch is to minimize fuel costs while satisfying constraints such as the power balance equation. The most fit individuals will have the lowest cost of the objective function of the economic dispatch problem. The fitness function is used to transform the cost function value into a measure of relative fitness. For the economic dispatch problem, the fitness function, Fit(P), may be expressed as

where popsize means population size and Pki is the generation power of the i-th unit at k-th chromosome. Reproduction involves creation of new offspring from the mating of two selected parents or mating pairs. It is thought that the crossover operator is mainly responsible for the global search property of the GA. We used an arithmetic crossover operator that defines a linear combination of two chromosomes [11]. Two chromosomes, selected randomly for crossover,

C igen and C gen j may produce two offspring,

+1 C igen +1 and C gen , which is a linear combination of their j

parents i.e.,

Fit ( P ) = g (ai + bi Pi + ci Pi 2 )
i =1

(8)

In order to produce two offspring, an arithmetic crossover operator is used. After crossover is completed, mutation is performed. In the mutation step, a random real value makes a random change in the m-th element of the chromosome. After mutation, all constraints are checked whether violated or not. If the solution has at least one constraint violated, a new random real value is used for finding a new value of the m-th element of the chromosome. Then, the best solution so far obtained in the search is retained and used in the following generation. The genetic algorithm process repeats until the specified maximum number of generations is reached. V. SIMULATION RESULTS The proposed method has been applied to two test systems; a power system with 6 units and a 20-unit system respectively. The cost function of each unit is chosen as a quadratic function for the test systems. The proposed method is compared with an improved Hopfield NN approach (IHN) [1], a fuzzy logic controlled genetic algorithm (FLCGA) [2], an advance engineered-conditioning genetic approach (AECGA) [3] and an advance Hopfield NN approach (AHNN) [4]. The proposed method was implemented in Matlab. The first test system has six units and details of this test system are given as follows: F1= 0.001562 P1 + 7.92 P1 + 561.0 F2= 0.00194 P22 + 7.85 P2 + 310.0 F3= 0.00482 P32 + 7.97 P3 + 78.0 F4= 0.00139 P42 + 7.06 P4 + 500.0 F5= 0.00184 P52 + 7.46 P5 + 295.0 F6= 0.00184 P62 + 7.46 P6 + 295.0
2

clear that the fuzzy logic controlled genetic algorithm [2] provides the worst results among the compared methods.
TABLE 2. RESULTS OBTAINED BY PGA, IHN, AHNN AND FLCGA Methods FLCGA IHN AHNN PGA Load (MW) 800.0 Cost ($/h) 8231.03 8228.05 8227.11 8227.09 Load (MW) 1200.0 Cost ($/h) 11480.03 11477.20 11477.09 11477.09

Table 3 gives a comparison of the economic dispatch results of AECGA, IHN, AHNN and PGA for 1520 MW and 2238 MW of load demand. It can be clearly seen that the proposed genetic algorithm achieved lower operation costs for different load demands, in contrast, the AECGA and the IHN produced higher operation costs for some loading conditions. For 2238 MW of load demand, the operation cost of the AECGA is 5.24 $/h higher than the operation cost of the proposed method. It is obvious that the proposed method produced the better solution than the compared methods for 6-unit system.
TABLE 3. RESULTS OBTAINED BY AECGA, IHN, AHNN AND PGA Methods AECGA IHN AHNN PGA Load (MW) 1520.0 Cost ($/h) 14169.54 14169.54 14169.54 14169.54 Load (MW) 2238.0 Cost ($/h) 20470.48 20465.44 20465.24 20465.24

100 < P1 < 600 100 < P2 < 400 50 < P3 < 200 140 < P4 < 590 110 < P5 < 440 110 < P6 < 440

In Table 1, the results of proposed method (PGA) are compared with the results of FLCGA, AECGA, IHN and AHNN when the load demand is 1800 MW. It is seen that there is a negligible difference in the values between the methods. The FLCGA produced a higher operation cost than other methods. Table 2 presents the results of PGA, IHN, AHNN and FLCGA when the load demands are 800 MW and 1200 MW. The FLCGA produced the highest operation cost and the obtained operation cost by the PGA are smaller than the IHN [1] respectively. For 800 MW of load demand, the operation cost of the AHNN is slightly higher than the operation cost of the proposed method. According to Table 1 and Table 2, it is

In order to demonstrate the efficiency and the robustness of the proposed genetic algorithm, a 20-unit system is considered. Details of this test system are obtained from [4]. The results of the proposed method is shown in Table 4 and compared against the results of the IHN and the AHNN for three load levels. The obtained operation costs by the IHN and the AHNN are about 0.74% and 0.2% higher than the PGA. The execution time of the proposed method for the 20-unit system is about 6 seconds.
TABLE 4. SIMULATION RESULTS FOR A 20-UNIT SYSTEM Operation Cost ($/h) Error % of IHN [1] 46460.84 53207.7 63791.72 AHNN [4] 46068.90 52981.76 63566.14 PGA 46027.35 52852.49 63401.84 IHN / AHNN 0. 94 / 0.09 0.67 / 0.24 0. 61 / 0.26

Load (MW) 3150 3800 4600

Methods FLCGA AECGA IHN AHNN PGA

TABLE 1. RESULTS OF FLCGA, AECGA, IHN, AHNN AND PGA FOR 6-UNIT SYSTEM Load Unit 1 Unit 2 Unit 3 Unit 4 Unit 5 Unit 6 (MW) (MW) (MW) (MW) (MW) (MW) (MW) 1800.0 1800.0 1800.0 1800.0 1800.0 250.49 248.07 248.08 248.14 248.14 215.43 217.73 217.74 217.74 217.74 109.92 75.30 75.18 75.20 75.20 572.84 587.70 587.90 587.80 587.80 325.66 335.60 335.55 335.56 335.56 325.66 335.60 335.55 335.56 335.56

Cost ($/h) 16585.85 16579.33 16579.33 16579.33 16579.33

VIII. BIOGRAPHIES VI. CONCLUSIONS In this paper, a new approach to the economic dispatch problem using a new genetic algorithm has been presented. The real valued representation scheme, arithmetic crossover, mutation and elitism are used in the genetic algorithms to generate successive sets of possible operating policies. In the case studies, the proposed method has been applied to the economic dispatch problem with 6 generators and 20 generators. Test results have shown the proposed genetic algorithm can provide better solutions than an improved Hopfield NN approach [1], a fuzzy logic controlled genetic algorithm [2], an advance engineered-conditioning genetic approach [3] and an advance Hopfield NN approach [4]. For a 6-unit system, the AECGA, the IHN and the AHNN produced higher operation costs than the new genetic algorithm for some loading conditions. The FLCGA produced the highest operation cost among the methods. The obtained operation costs by the IHN [1] and the AHNN [4] are about 0.74% and 0.2% higher than the PGA for a 20-unit system.
Tankut Yalcinoz (S'1991, M'2000) was born in Mersin, Turkey, in 1968. He received the B.Sc. degree in electrical and electronic engineering from Karadeniz Technical University, Turkey in 1990 and the M. Sc. degree from Cukurova University, Turkey in 1993. In 1998 he obtained a Ph.D. from Imperial College, London. Since 1998 he has been a Lecturer at Nigde University. His main interests are in power system planning and operation and the application of computational intelligence to power systems. Halis Altun (M'1996) was born in Nevsehir, Turkey, in 1968. He recieved the BSc degree in Electronic and Communication Engineering from Yldz University in 1991. In 1998 he obtained a Ph.D. from Univeristy of Nottingham, UK. Since 1998 he has been a lecturer at Nigde University, Turkey. His main interests are in neural networks, signal and speech processing. Murat Uzam (M'1998) was born in Aydin, Turkey, in 1968. He received both the B.Sc. and M.Sc. degrees in electrical engineering from Yildiz Technical University, Turkey in 1989 and 1991 respectively. In 1998 he obtained a Ph.D. from Salford University, UK. Since 1998 he has been a Lecturer at Nigde University. His interests include modeling and control of discrete event systems using Petri net based techniques.

VII. REFERENCES
[1] [2] [3] [4] T. Yalcinoz and M.J. Short, "Large-scale economic dispatch using an improved Hopfield neural network," IEE Proc. Gener. Transm. Distrib., vol. 144, pp. 181-185, March 1997 Y.H. Song, G.S. Wang, P.Y. Wang, A.T. Johns, "Environmental/ economic dispatch using fuzzy logic controlled genetic algorithms," IEE Proc. Gener. Transm. Distrib., vol. 144, pp. 377-382, July 1997. Y.H. Song and C.S.V. Chou, "Advanced engineered- conditioning genetic approach to power economic dispatch," IEE Proc. Gener. Transm. Distrib., vol. 144, pp. 285-292, May 1997. T. Yalcinoz and H. Altun, "Comparison of simulation algorithms for the Hopfield neural network: an application of economic dispatch," Turkish Journal of Electrical Engineering & Computer Sciences, vol. 8, no.1, pp. 67-80, 2000. H.H. Happ, "Optimal power dispatch A comprehensive survey," IEEE Trans. Power App. Syst., vol. PAS-96, pp. 841-854, 1977 B. H. Chowdhury and S. Rahman, "A review of recent advances in economic dispatch," IEEE Trans. on Power Systems, vol. 5, pp. 12481259, November 1990. G.B. Sheble and K. Brittig, "Refined genetic algorithm - economic dispatch example," IEEE Trans. on Power Systems, vol. 10, no. 1, pp. 117-124, 1995. D. Walters and G.B. Sheble, "Genetic algorithm solution of economic dispatch with valve point loading," IEEE Trans. on Power Systems, vol 8, no. 3, pp. 1325-1331, 1993. A. Bakirtzis, V. Petrides and S. Kazarlis, "Genetic algorithm solution to the economic dispatch problem," IEE Proc. Gener. Transm. Distrib., vol. 141, no. 4, pp. 377-382, July 1994. P.H. Chen and H.C. Chang, "Large-scale economic dispatch approach by genetic algorithm," IEEE Trans. on Power Systems, vol. 10, no. 4, pp. 1919-1926, November 1995. Z. Michalewicz, Genetic Algorithms + Data Structures = Evolution Programs, 2nd ed., Springer-Verlag, Berlin, 1994. A.J. Wood and B.F. Wollenberg, Power Generation Operation and Control, John Wiley & Sons, New York, 1984. A.H. Wright, Genetic algorithms for real parameter optimization, in Foundations of Genetic Algorithms, J.E. Rawlins (Ed.), Morgan Kaufmann, pp. 205-218, 1991.

[5] [6] [7] [8] [9] [10] [11] [12] [13]